Are AGI timelines extrapolating the wrong axis?

Most AGI timelines extrapolate capability curves, which is reasonable if you believe the current architecture can in principle reach general intelligence by scaling. There is a different reading in which the architecture is good at one thing and bad at another, and what AGI actually requires lives on the bad side. The good side is computation inside a delineated frame. The bad side is moving across frames, recognizing when the world has shifted, and reorienting. Calling both intelligence runs them together.